Community Research and Development Information Service - CORDIS

Abstract

The aim of this research is to demonstrate the possibility of automatic recognition of surface defects in cold flat steel products. The existing systems detect the defects but give poor results in identification. Defect identification in our application has three main advantages:

- To disminish operator workload by presenting only high and medium severity defects. This needs real time processing;
- To facilitate the choice in product destination based on the overall quality of the coil;
- To detect and identify process malfunctioning.

The following results have been obtained:

- Definition of the global architecture of the automatic recognition systems. This study shows the need for a primary image texture analysis;
- Bibliography on automatic classification;
- Classification trials on a first image database, manually segmented;
- Development of the image texture analysis and classification method;
- Development of an automatic segmentation method for the main kind of texture images (80% of production);
- Classification tests on a more complete database, segmented using the automatic segmentation method;
- Definition of a quality criterion with a cost matrix allowing choice of the classifier with an iterative classifier building method, in order to reduce the number of misclassified high severity defects.
ty defects.

Additional information

Authors: ODET C, Institut national des sciences appliquées, Villeurbanne (FR);DUPONT F, Institut national des sciences appliquées, Villeurbanne (FR);GOUTTE R, Institut national des sciences appliquées, Villeurbanne (FR)
Bibliographic Reference: EUR 16661 FR (1997) 145pp., FS, ECU 25
Availability: Available from the (2)
ISBN: ISBN 92-827-9300-1
Record Number: 199710762 / Last updated on: 1997-06-23
Category: PUBLICATION
Original language: fr
Available languages: fr